Search

The authors describe a LISP microprocessor which includes over 550 K transistors, has 114 K of on-chip RAM, and runs instructions in a single 30-ns clock cycle. The chip is implemented in 1.25-/spl mu/m double-level-metal (DLM) CMOS, has 224 pins, and is packaged in a custom pin-grid array. The microinstruction and macroinstruction sets of this chip are compatible with an existing LISP processor. An extensive discussion of test features designed into the processor chip is given.

Circuit simulation is a very time-consuming and numerically intensive application, especially when the problem size is large as in the case of VLSI circuits. To improve the performance of circuit simulators without sacrificing accuracy, a variety of parallel processing algorithms have been investigated. Research in the field of parallel circuit simulation is surveyed, and the ongoing research in this area at the University of Illinois is described. Both standard and relaxation-based approaches are considered. In particular, the forms of parallelism available within the direct method approach, used in programs such as SPICE2 and SLATE, and within the relaxation-based approaches, such as waveform relaxation, iterated timing analysis, and waveform-relaxation-Newton, are described. The specific implementation issues addressed are primarily related to general-purpose multiprocessors with a shared-memory architecture having a limited number of processors, although many of the comments apply to a number of other architectures. >

Several physical phenomena in highly scaled CMOS technology have now become first order elements affecting electrical behavior of transistor characteristics. Effects such as STI mechanical stress, direct tunneling in gate dielectrics, gate line-edge roughness, and others, have significant influence on device characteristics. This paper elaborates on these effects to exemplify the need for closer interaction between circuit design and process development teams in order to push out application-dependent scaling limits. The paper also highlights the need for further efforts in the areas of circuit-level device modeling.